Need help with system-upgrade-controller?
Click the “chat” button below for chat support from the developer who created it, or find similar developers for support.

About the developer

158 Stars 25 Forks Apache License 2.0 88 Commits 15 Opened issues


In your Kubernetes, upgrading your nodes

Services available


Need anything else?

Contributors list

System Upgrade Controller


This project aims to provide a general-purpose, Kubernetes-native upgrade controller (for nodes). It introduces a new CRD, the Plan, for defining any and all of your upgrade policies/requirements. A Plan is an outstanding intent to mutate nodes in your cluster. For up-to-date details on defining a plan please review v1/types.go.


Presentations and Recordings

April 14, 2020

CNCF Member Webinar: Declarative Host Upgrades From Within Kubernetes - Slides - Video

March 4, 2020

Rancher Online Meetup: Automating K3s Cluster Upgrades - Video


Purporting to support general-purpose node upgrades (essentially, arbitrary mutations) this controller attempts minimal imposition of opinion. Our design constraints, such as they are:

  • content delivery via container image a.k.a. container command pattern
  • operator-overridable command(s)
  • a very privileged job/pod/container:
    • host IPC, NET, and PID
    • host root file-system mounted at
  • optional opt-in/opt-out via node labels
  • optional cordon/drain a la

Additionally, one should take care when defining upgrades by ensuring that such are idempotent--there be dragons.


The most up-to-date manifest is always manifests/system-upgrade-controller.yaml but since release v0.4.0 a manifest specific to the release has been created and uploaded to the release artifacts page. See releases/download/v0.4.0/system-upgrade-controller.yaml

But in the time-honored tradition of

curl ${script} | sudo sh -
here is a nice one-liner:

```shell script


kustomize build | kubectl apply -f - ```

Example Plans

Below is an example Plan developed for k3OS that implements something like an

of content from the container image to the host, preceded by a remount if necessary, immediately followed by a reboot.
kind: Plan


This name should be short but descriptive.

name: k3os-latest

The same namespace as is used for the system-upgrade-controller Deployment.

namespace: k3os-system


The maximum number of concurrent nodes to apply this update on.

concurrency: 1

The value for channel is assumed to be a URL that returns HTTP 302 with the last path element of the value

returned in the Location header assumed to be an image tag (after munging "+" to "-").


Providing a value for version will prevent polling/resolution of the channel if specified.

version: v0.10.0

Select which nodes this plan can be applied to.

nodeSelector: matchExpressions: # This limits application of this upgrade only to nodes that have opted in by applying this label. # Additionally, a value of disabled for this label on a node will cause the controller to skip over the node. # NOTICE THAT THE NAME PORTION OF THIS LABEL MATCHES THE PLAN NAME. This is related to the fact that the # system-upgrade-controller will tag the node with this very label having the value of the applied plan.status.latestHash. - {key:, operator: Exists} # This label is set by k3OS, therefore a node without it should not apply this upgrade. - {key:, operator: Exists} # Additionally, do not attempt to upgrade nodes booted from "live" CDROM. - {key:, operator: NotIn, values: ["live"]}

The service account for the pod to use. As with normal pods, if not specified the default service account from the namespace will be assigned.


serviceAccountName: k3os-upgrade

Specify which node taints should be tolerated by pods applying the upgrade.

Anything specified here is appended to the default of:

- {key:, effect: NoSchedule, operator: Exists}


  • {key:, effect: NoSchedule, operator: Equal, value: amd64}

  • {key:, effect: NoSchedule, operator: Equal, value: arm64}

  • {key:, effect: NoSchedule, operator: Equal, value: arm}

    The prepare init container, if specified, is run before cordon/drain which is run before the upgrade container.

    Shares the same format as the upgrade container.


    If not present, the tag portion of the image will be the value from .status.latestVersion a.k.a. the resolved version for this plan.

    image: alpine:3.11 command: [sh, -c] args: ["echo '### ENV ###'; env | sort; echo '### RUN ###'; find /run/system-upgrade | sort"]

    If left unspecified, no drain will be performed.





    deleteLocalData: true # default

    ignoreDaemonSets: true # default

    force: true

    Use disableEviction == true and/or skipWaitForDeleteTimeout > 0 to prevent upgrades from hanging on small clusters.

    disableEviction: false # default, only available with kubectl >= 1.18

    skipWaitForDeleteTimeout: 0 # default, only available with kubectl >= 1.18

    If drain is specified, the value for cordon is ignored.

    If neither drain nor cordon are specified and the node is marked as schedulable=false it will not be marked as schedulable=true when the apply job completes.

    cordon: true


    If not present, the tag portion of the image will be the value from .status.latestVersion a.k.a. the resolved version for this plan.

    image: rancher/k3os command: [k3os, --debug]

    It is safe to specify --kernel on overlay installations as the destination path will not exist and so the

    upgrade of the kernel component will be skipped (with a warning in the log).


    • upgrade
    • --kernel
    • --rootfs
    • --remount
    • --sync
    • --reboot
    • --lock-file=/host/run/k3os/upgrade.lock
    • --source=/k3os/system
    • --destination=/host/k3os/system


shell script




Also see

that spells out what a "typical" deployment might look like with default environment variables that parameterize various operational aspects of the controller and the resources spawned by it.


Integration tests are bundled as a Sonobuoy plugin that expects to be run within a pod. To verify locally:

shell script
make e2e

This will, via Dapper, stand up a local cluster (using docker-compose) and then run the Sonobuoy plugin against/within it. The Sonobuoy results are parsed and a

Status: passed
results in a clean exit, whereas
Status: failed
exits non-zero.

Alternatively, if you have a working cluster and Sonobuoy installation, provided you've pushed the images (consider building with something like

make REPO=dweomer TAG=dev
), then you can run the e2e tests thusly:

shell script
sonobuoy run --plugin dist/artifacts/system-upgrade-controller-e2e-tests.yaml --wait
sonobuoy results $(sonobuoy retrieve)


Copyright (c) 2019-2020 Rancher Labs, Inc.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.